Fuzzy based approach for detection of object in space-based observation

Abstract

The study of space weather is an integral aspect of modern day’s space-based and ground-based technological systems. High-speed solar wind stream (HSSs) is one of the main drivers of space weather, resulting in severe geomagnetic storms and affect navigation, communication and low Earth-orbiting satellites as-well-as high latitude electrical grids and cause aurora in the polar atmosphere. Solar coronal hole (CH) is one of the important sources of these HSSs. Therefore, proper detection of solar CH is one of the key ingredients for the successful prediction of space weather conditions. CH can be identified from space-based solar observations in X-rays and extreme ultraviolet (EUV) rays. However, the CH region is having a relatively low signal-to-noise ratio (SNR). Alongside, CH detection suffer from challenges like, increased vagueness, uncertainty and impreciseness and blur region boundary. The detection becomes more complicated if the object of interest is situated within another object in the image, as observed in the case of solar coronal holes (CHs) and filament detection. In this thesis, fuzzy-based single and dual contour object detection approaches have been developed for automated and fast coronal hole detection. The developed methods have been applied on solar images captured by space-based SDO/AIA instrument. Comparisons have been carried out with the existing state of-the-art methods for quantitative and qualitative analysis. It is observed that the proposed fuzzy energy-based dual contours model (FEDCM) detects the CHs regions with highest F1 score (Dice coefficient) of 0.7463 ± 0.1165 compared to the existing active contour without edge (ACWE), convolutional neural network (CNN), spatial possibilistic clustering algorithm (SPoCA) and coronal hole identification via a multi-thermal emission recognition algorithm (CHIMERA) based method which are having Dice coefficient (0.5959 ± 0.1013), (0.5656 ± 0.1393), (0.5157 ± 0.1469) and (0.6921 ± 0.1362) respectively, and generate results within a time period of (57.39 ± 2.72) sec which is quite lower than existing ACWE, CNN and CHIMERA methods having the execution time (140.62 ± 1.07) sec, (239.90 ± 1.20) sec and (99.74±19.29) sec respectively . Whereas, the developed parameterized online region based active contour method (POR-ACM) segment the CHs with Dice coefficient rate of 0.7090 ± 0.1076 within (89.96 ± 15.58) sec. At the same time, among the proposed methods the fast fuzzy c-means (FFCM) clustering followed by the circular Hough transform (CHT) simulated static contour method (FFCM-SCM) having the less execution time period of (51.84 ± 6.14) sec which is slightly higher compared to existing SPoCA’s execution time of (49.21 ± 5.38) sec, with Dice coefficient rate of 0.6581 ± 0.1171.